Carousal Starts

Generative AI
Handbook
What to know and what actions to take to prepare
your business for an era of exponential innovation

Carousal Ends

Landscape Basics Applications Responsible AI Readiness

How we arrived here, The strengths, Known and Deploying gen AI Practical guidance
where we may be weaknesses and expected use -cases solutions in a for adapting and
headed and meeting unknowns of the of gen AI in responsible and leading in a gen AI
the natives technology enterprise effective way world
environments

Generative AI will reshape every industry. But when?

There’s little question that gen AI has captivated business interest since ChatGPT launched at the end of 2022. Interest has only
grown since that announcement and we believe it will transform organizations through new levels of human-machine
collaboration. Throughout this guide you’ll find statistics, predictions and perspectives to spur thinking on how to pragmatically
apply this technology to innovate.

The sense of urgency is understandable given rapid advancements. However, while most companies have actively explored gen
AI’s potential through proofs of concept and early-stage experimentation this past year, Cognizant research shows that many
leaders (30%) believe meaningful impact is still years away.

For most executives we engage, the question is not “if” but “how and when” gen AI will transform their business models and
operations. Many are now looking to scale early successes through broader initiatives. Our own research and client
conversations this past year reveal enthusiastic curiosity tempered by thoughtful diligence around these emerging capabilities.
As enterprises look to transition experiments into scaled production-grade solutions, understandable caution accompanies the
excitement.

Still, through skills-building and laying responsible foundations in 2023, companies equipped themselves for the next stage of
maturity in leveraging AI’s generative potential. 2024 is the year to accelerate AI impact through focus and investment. The rules
of engagement continue to rapidly evolve as practical experience refines our thinking on the possible. By working together, we
can apply this technology practically and responsibly to increase productivity and deliver superior human-centric experiences.

Let’s shape the future together with care, creativity and purpose.

Carousal Starts

Landscape
The human-like ability of generative AI to converse,
consider and create has captured imaginations. By
understanding how we got here—and the decades of
thinking that led us to gen AI—we can better predict
what’s coming next.

Carousal Ends

How we arrived here

“Generative AI sits
An innovation boom 75 years in the making within the context of
decades of research
In the summer of 2022—well before Open AI’s ChatGPT exploded into the public
discourse—gen AI began generating media buzz when a recently-fired Google into AI. We are now
Engineer claimed1 their LaMDA model might be sentient. entering the era where
this technology will start
True or not, this wasn’t an entirely surprising claim for artificial intelligence.
We’ve been expecting this. to fundamentally
transform businesses.”
Since Alan Turing’s 1950 “Imitation Game” (Turing Test) proposal, we’ve imagined
a future of computers with human-like intelligence, personality and autonomy. Naveen Sharma
Today, we seem to be accelerating towards that disruptive future.
Head of Artificial Intelligence
If and when we reach sentient AI is anybody’s guess but gen AI presents a and Analytics Practice
significant new milestone in this journey, sparking new interest, innovation and
discourse. Here’s how we arrived at this moment.

1950s – 60s 1970s – Mid-90s Mid-90s – 2000s 2010s - Present

The birth of AI The two AI winters The rise of machine learning The Advent of Deep Learning

In the early days of modern While progressive advances With the internet and Significant breakthroughs in
computing, the Turing Test like expert systems proliferate accelerated business neural network and
and the famed 1956 and cultural interest in AI digitization, data availability generative AI model
Dartmouth Workshop launch grows, underwhelming results and IT funding expand to development, accomplishing
the era of AI. The first AI lead to two periods of drive practical AI applications. previously impossible tasks,
programs form and optimism reduced funding and focus. Major advances in robotics alongside surge in big-tech
is high. and data-driven solutions. investment. As of Q1 2024, the
Crunchbase AI startup list has
grown to nearly 10,000
companies2.

Where we’re headed

Change (and more change) ahead

In an April 2023 interview with CBS, Alphabet

CEO Sundar Pichai famously commented that
today’s generative AI will soon impact “every
product across every company.”3 With rapid
consumer adoption and the increased
competitive pressure created by gen AI across
industries, Pichai’s predication is sure to hold
true.

As gen AI permeates markets, it’s critical that

adaptability be built into the technology and
cultural fabric of organizations. New, disruptive
intra-industry and extra-industry use-cases will
arise frequently in the coming years creating
continuous change to navigate.

1 2 3 4 5
A much larger context New gen AI models, Waves of regulation Generative video and War for talent shifts
window expanded AI features in and standards AR/VR renaissance to war for innovation
Increasing context enterprise software World governments With significant As 30% of work
windows are critical for Next-gen models are will adopt and adapt advancement in hours4 are expected
many enterprise use- already in development, regulations at lagging AR/VR technology to be directly
cases and will allow for including open-source pace as they address spearheaded by Meta, impacted by AI and
larger, more models with more rapidly evolving Apple and Microsoft, resulting automation
comprehensive prompts flexibility and control. ethical, economic and compelling new capabilities,
to be passed to models. Expect acceleration of societal concerns. applications backed productivity gains
This new access to vast new entrants and Organizations will by gen AI will launch. will be felt by all. The
contextual datasets will innovation. Enterprise formalize AI With conversational war for technology
open even more doors platforms are adding AI governance roles with user interfaces (i.e., talent will be
for AI. tooling that will drive variable risk tolerance chat, voice), new visual reshaped as a war
further proliferation. for use cases. worlds will be seen. for technology
innovation as
organizations
differentiate with
data.

Meeting the AI natives

The forever digitally young

Operating effectively in the era of generative AI requires a reconstruction of the now decades-old digital maturity narrative.
We’re entering a post-digital era where every enterprise is digital and what defines leaders is their adaptability—which extends
to their definition of maturity, how they operate and what they sell.

The gen AI era is one of continuous, perpetual change. The fundamental definition of computation and chip design is changing
with AI as rigid, linear and exact computations are swapped for the abstract and inexact logic that underpins neural network
thinking. Companies must now operate in similar ways.

So, are there AI natives?

As noted in our gen AI timeline, there has been an explosion of AI-centric startups born over the past two years—these might be
defined as AI natives. These companies focus on AI and, presumably, they have AI built into their operations and culture as well
as their product.

But just as our definition of digital maturity requires a ‘continuous change’ perspective, so too will our definition of the “AI-native
company”.

Being “born into” the gen AI era is far less important than exploration and adoption. Those organizations who pioneer AI—and
set the rules early to gain competitive market share from it—will establish what it means to be an AI native. Enterprise
organizations, with their robust proprietary data to build upon, have the advantage.

Early pioneers have common traits:

• Modern operations and architecture: Delivery and IT infrastructures that leverage data and accelerate change
• Enabled AI culture: Definitive policies and governance with access and training for staff to benefit from AI
• A focus on disruptive propositions: Already piloting AI experiences to hedge against new entrant threats
• Engaging suppliers to evolve services: Addressing gen AI opportunities for mutual benefit
• Simplifying not complexifying: See gen AI as a way to transform value creation vs. ”just new tech”

How are early pioneers applying generative AI?

• Expert assistants for professional work

Gen AI assistants being used in legal software to augment services and accelerate work

• Media generation
Entertainment and creative studios using gen AI to create animation sequences, social reels and more

• Market insights
Aggregating customer feedback (reviews, call transcripts, etc.) for gen AI-driven insights

• Text to software product generators

Gen AI using component libraries, design systems and code bases to build POC software “ondemand”

• Interactive chat-bots
Transforming the classic chat-bot, gen AI powering conversational agents to answer and explain

Carousal Starts

Basics
The fundamental strengths of generative AI perfectly
mirror its unavoidable weaknesses. The fundamental
characteristics of the technology provide insight into
its disruptive potential – and explain why adoption will
impact every part of the enterprise over time.

Carousal Ends

Known strengths of generative AI

“Generative AI provides
The blurring boundary between human and machine work
completely new
Gen AI presents a fundamental change in our understanding of what practical, capabilities to
immediately-accessible AI can do. Chat-bots, candidate screening tools, automate and
summarizers and picture-makers might inspire us today, but soon AI will shape the
core of modern business.
augment knowledge
work. It is going to
It isn’t sentient but it sure does behave in human ways – and that’s what’s so turbo-charge tasks that
inspiring about this technology. Whether finishing a sentence, writing the code for
a component, ideating on novel molecular structures or animating an entire new
require creativity and
movie, this generation of AI composes complex patterns and data to create. expertise, such as
design, engineering
As organizations come to understand the strengths and potential use-cases of
and quality assurance”
gen AI, they also begin to realize the fundamental requirements within their
organization for fully leveraging this technology. Pramod Bijani
It all starts with data. Combine that data with business processes and logic and Head of Digital Experience and
gen AI is transformative. Digital Engineering Delivery

1 2 3 4 5 6
Complex process Data Predictive analysis Driving efficiency Real time Multimedia
automation augmentation and Gen AI is capable and supporting optimization generation
Core business completion of analyzing knowledge work Gen AI is capable Gen AI is
processes that, in Gen AI systems complex, AI makes of monitoring capable of both
the past, have not work with data to structured or knowledge work processes and consumption
been open to provide first-line unstructured data more efficiently by outputs to and creation of
automation due to analysis, to identify patterns accelerating and proactively identify rich media
complexity and classification, and trends to form expanding on opportunities for across text,
variability can now sanitization and actionable ideation, distilling improvement, audio, video and
be managed and more, free from recommendations data to find prescribe and even images,
reshaped by AI human error and insights, rapidly implement unlocking
at scale drafting and more changes powerful new
possibilities

Known weak points of generative AI

To err is human. For AI, it’s by design.
Hallucinations
The cyclical evolution of AI over the past 75 years has been
marked by periods of waxing enthusiasm and waning When AI produces unreliable and erroneous outputs, it
pessimism. As new advances promised new opportunities, erodes data-driven strategy, reduces customer trust and
institutions and businesses have jumped in and invested limits operational efficiency.
heavily in the technology. When outcomes haven’t met
expectations, though, the AI space has experienced
disillusionment and stagnation.

By its nature, AI is unpredictable and this is only further

Data quality and AI safety
highlighted by the gen AI boom and the human-like creativity Because data shapes AI’s knowledge base, any
of its outputs. inadequate data inputs will create bias and limit accuracy,
fairness and decision-making.
Like humans and on many tasks, gen AI is capable of working
flexibly towards a goal or target output rapidly and creatively.
Also, like human workers, context matters.

Whether its brand values, ethical considerations, situational The explainability problem
knowledge, historical learning, consumer needs or anything
else, human workers are expected to understand the context of When AI provides a decision, there is inherent uncertainty
their work—and this can impact the output of their efforts. in its certainty. The larger and more complex a model, the
With generative AI, contextual understanding is often difficult harder to ascertain.
to achieve “out of the box,” especially with consumer tools like
ChatGPT. This has been the source of much of gen AI’s
criticism.

Its strengths shape its weaknesses

“Overcoming gen AI’s limitations, we
Generative AI wants to answer us. It’s built to respond to our
prompts—no matter their complexity—and often provides
are achieving very strong results in
answers that, in a sense, acknowledge this fact. ChatGPT hybrid systems where generative and
allows us to refresh responses. Image generators like OpenAI’s evolutionary AI models are combined
DALL-E or the popular Midjourney both return multiple images to play to their strengths. This will be a
to any single prompt. These tools understand that they may be
wrong.
critical foundation for successful
adoption.”
The creative muscle of gen AI can be striking, but it is definitely
not magic. Gen AI’s capabilities are fundamentally based on
Babak Hodjat
reference data and training.
AI CTO
AI adoption creates new categories of risk that require focused
assurance at the enterprise level. Organizations that engage in
this transformative technology with this in mind will gain the
most from the AI era.

Gen AI’s open questions

Waiting is not the answer

As enterprise decision-makers consider their

paths forward in this new AI-enabled world,
questions are being asked that are as complex
and difficult to navigate as the technology
itself. How do we know our AI isn’t
hallucinating? What ethical boundaries does
this system present? How can we rely on
answers, if we can’t explain how our systems
reached them?

These questions largely mirror the ‘weaknesses’

discussed earlier, and their answers will be
shaped through trial and error, learning and
time.

What’s certain is that readying the organization

to navigate this AI-enabled world is critical for
future business performance—exploring these
questions is a key part of that readiness.

We’ll explore near-term steps to take to address

these concerns within the readiness section
later in this guide.

How can we ensure ethical What legal implications How can we ensure How will this impact our
usage? should concern us? predictable output? brand or public perception?

Significant concerns for Concerns for copyright, IP Concern over Concern over customer
misuse and harm, bias in infringement and regulatory hallucinations, poorly preference against gen AI in
poorly trained systems and issues when dealing with structured or insufficient key touchpoints and dilution
other negative outcomes of protected data, privacy direction in business- of brand experience
use impacting scenarios
Cognizant answer: These Cognizant answer: As with
Cognizant answer: AI concerns are ‘native’ to Cognizant answer: other AI questions, the key is
creates entire new digital business. Processes Generative AI can never be to test and train gen AI
categories of risk that and tooling can help expected to provide 100% systems well before usage
require focused testing, navigate and safeguard predictability in output. This and be selectively
governance and assurance. here and every business is why training, tuning and transparent with customers
LLMOps creates a needs a unique plan continuous monitoring must about use
framework for such be integrated
responsible usage

Carousal Starts

Applications
Consider the early plugins available for ChatGPT, or
bots on the Poe app, and it’s clear that the use -cases
of generative AI are about as vast and varied as
software itself—and those are just chat interfaces.

Carousal Ends

1 2 3 4 5 6 7
Market and Software Production Marketing Expert Employee Customer
competitor development and sales advisors engagement experience
intel

7 proven use cases of gen AI impact

Even at this early stage, the opportunities for generative Al across the enterprise are countless. With the right foundations, the
only limitation of gen AI solution-building may be a company’s imagination.

With so much opportunity and so many questions, it can be hard to know where to start. As you’ll find in our discussion of gen AI
readiness later in this guide, what’s key is that organizations begin exploring this technology early to identify their own
opportunity spaces, safeguard against disruption and begin building skills.

Fortunately, there’s no need to start this journey from scratch. With the following seven example use-cases of generative AI, we’ll
highlight just how varied the opportunity can be. Every part of the value chain across every industry stands to be disrupted in
unique, differentiating ways as organizations bring their unique data, processes and POV to the discussion. Let’s dig in.

1. Market and competitive intel

Turning data into human-readable, actionable and contextualized guidance is a major strength of gen AI. Generative AI systems
can be used to industrialize data collection from a range of sources, including curated market research, real-time customer and
competitive behavior, internet scraping and primary user research. Whether structured or unstructured, this data empowers
systems to drive a range of automated analysis, summarization and recommendations.

Ingestion of market data Gaining insights from data Developing and validating Packaging and distributing
hypotheses

Summarize and classify Synthesize multiple sources Concept mock-up Report writing
Collate captured data to Correlate, compare and Mock up concepts and Generate narrative and
identify relevance to combine related content on stimulus for research testing explanation around raw
defined topics of interest a topic(s) to distill insights and user validation data and derived insights

Coverage adjustment Ideate Transcription Insight navigator

Identify new, relevant topics Respond to insights to Transcribe research data, Provide a natural language
and add granularity to generate new concepts and extracting findings, discussion interface to
existing topics being hypotheses confidence levels and interact with the corpus of
tracked aspects requiring follow-up research data
Prepare primary research research
React as persona Define audiences and
Adopt customer and create qualitative and
stakeholder personas and quantitative research
react to content questions

2. Software development
Microsoft Visual Studio Code, the wildly popular integrated development environment (IDE), has long-supported GitHub’s Copilot
product (by some estimates automating 40%-60% of code writing5) and now also integrates ChatGPT directly into the developer
interface. But the utility of generative AI during software development goes well beyond writing components. The entire software
development process is set to see transformation as this technology impacts creativity, quality, productivity, compliance, utility
and more.

Generate ideas Develop concepts Implement software Assurance and release

Crowd sourced ideation Concept mock-up Code completion Release readiness

Collate employee ideas, Create rapid product Create code based on Inspect code and project
feedback and data to visualizations to bring to life collaborative developer actions to determine if go-
identify new opportunities new ideas and support user input and code base no-go criteria are met and
testing context the solution is ready to
Navigate current state release
Engage with AI to discuss Requirements elaboration Quality assurance
enterprise structure, Extrapolate concepts to Assess risks and failure
performance, code base, epics, stories and modes, build test scenarios,
etc. acceptance criteria test data and automations
to validate
User testing Implementation planning
Use AI for scope testing Prioritize scope items, Change management
sessions, to summarize sequence to sprint plans Identify impacted groups
findings or suggest and assign to teams based and assist those groups in
improvements on functions and skills addressing impacts

3. Production
Product research, production and quality control will see significant Generative AI impact in the coming years as organizations
across industries seek to unlock transformative new efficiency and product innovation ahead of competition. This zone is highly
controlled and data-intensive, making it a perfect early adoption area. The IP established through smartly leveraging Generative
AI in this space will reshape industries and establish new leaders.

Research and development Design and prototyping Production Go to market

PRD shaping Generative design Cost/sustainability External systems integration

Shape product Create numerous variants optimization Gen AI can be integrated
requirements based on based on requirements, Analyze design-to- with ERPs and CRMs for
current and expected future accelerate creative work production pipeline for cost further insights
demands savings and sustainability
Tradeoff-based prototyping Continuous improvement
Materials research Rapid virtual prototypes Production quality feedback loop
Assist in identifying and based on adjustable goals assurance Analyze product
optimizing materials for a and balance of trade-offs Systems to inspect performance data and
variety of requirements products, ID defects or feedback to drive
Niche user optimization inconsistencies accurately refinements
AI-powered scenario Simulate specific user at scale
modeling scenarios, suggest variants
Emulate real-world that service new groups Supply chain optimization
scenarios virtually at vast Predict supply chain
scale to optimize product Design critique AI disruptions, optimize
design Partner with humans to sourcing and optimize
critically respond to logistics
concepts and help shape
design

4. Marketing and sales

Generative AI improves planning, production efficiency and effectiveness throughout the marketing and sales journey. As the
technology gains adoption, asset production cycles will see a marked acceleration with a range of potential new asset types
and channel strategies becoming available. Further, self-service channels will become more personalized and impactful while
sales staff will increase their productivity and knowledge to focus more time on driving successful customer engagements.

Manage channels Nurture leads Manage opportunities Negotiate and close

Content production Lead capture Sales effectiveness Customer RFP responses

Create digital assets: Conversational data Train sales staff on latest Respond automatically
images, alt text, copy, capture at lead propositions, messaging using best practice answers
translations, scripts, identification stage to and tactics tailored to context
campaign plans and more provide rich lead
information and immediate Sales presentations Negotiation and closure
Software development customer response Generatively compose Assemble contracts and
Create software to support presentations that combine statements of work to
interactive campaign Lead nurturing sales strategy, customer match proposals, MSAs,
elements Create personalized context and standard standard terms, etc.
comms, content and offerings
User testing experiences for
Scope testing sessions, engagement and nurturing
summarize findings and leads
suggest improvements
Sales prioritization
Identify the best activities
for sales teams to focus on
and craft personalised
action plans

5. Expert advisors
Generative AI streamlines and accelerates the provisioning of expert advice to benefit end-users and businesses alike. In many
scenarios, gen AI has the capacity to act in a self-service model to provide expert guidance directly to users. Where complexity
is higher or in safety-critical environments, gen AI can facilitate many stages of the process without acting in a fully autonomous
way. With AI-driven pre- and post-processing, experts can more effectively utilize their time and focus on the highest-value or
most-critical scenarios.

Customer requests service Collect information Evaluate and recommend Action recommendations

Qualification Information extraction Anomaly extraction Downstream fulfilment

Understand user issue and Process unstructured data Draw attention to Interpret expert conclusions
determine if expert sources to extract the interesting or outlier input and trigger appropriate
consultation is needed and required information data features follow-up activities
appropriate
Initial data capture Recommendation Quality assurance
Triage Address questions in a Apply standard policies and Review expert conclusions
Understand the urgency conversational experience procedures to the input and highlight any causes
and importance of the to gather necessary data context to suggest for concern (e.g., policy
request to prioritize service outcomes misalignment, bias)
Collection management
Chase participants to Case discussion
provide missing information Allow experts to
and correct errors interactively discuss the
case with the necessary
background knowledge to
take action

6. Employee engagement
Companies that adopt generative AI at a cultural level, going beyond asset production and chat interactions to elevate all
common touch-points for customers and employees alike, will see the biggest gains in the coming years. Employee engagement
is an exciting space for gen AI with the potential to impact recruiting, onboarding, team-building, performance management,
support and more. The efficiency gains here will empower innovation across the business as gen AI permeates the market.

Recruit Onboard Enable Develop

Candidate screening New hire buddy Accelerate admin Performance coach

Review applications, assess Conversational onboarding Simplify standard activities Review activities, outcomes
fit and extract concerns assistant to guide employee like timesheets, vacation and feedback to suggest
prior to interview journeys for first month requests, expense reporting, specific learning and
training, etc. development activities
Interview assistance Staffing
Suggest role-specific Leverage enriched findings Support AI
questions, transcribe and to match new hires to roles Assist with self-service
summarize conversations and opportunities support for IT, human
resources, etc.
Close Community connection
Develop tailored offer Suggest connection to Educational AI
strategy to close strategy communities of interest, Provide interactive training
using interview transcripts sites, etc. on skills development topics
and answer questions on
company data

7. Customer experience
Whether they’re just browsing or already a loyal customer, the way that people engage with brands throughout the shopping
and post-purchase experience is set to dramatically evolve with gen AI. With answers becoming more seamless and appetite for
content noise decreasing, customers will expect personal, intuitive, adaptive touch-points that understand and serve their needs.
It’s time to put more you into UX.

Shopping and product Purchase, returns Loyalty Repeat

comparison

Intelligent product Dynamic personalized Generative loyalty Behavioral campaign

suggester pricing programs shaping
Synthesize wide user data Potential for novel AI-driven Novel loyalty programs Gen AI learns from loyal
to create a new level of pricing mechanisms based using gen AI systems that buyers to identify/surface
personalized shopping on various inputs to personalize incentives trends and formulate full-
experience optimize sales funnel campaigns
Churn prevention AI
Virtual demos / try-ons Predictive return systems ID behavioral patterns of at-
Generative AI visualizes Based on prior buying risk customers and design
product in various settings, behavior, post-purchase engagements to prevent
including worn, for better engagements, etc. predict churn
product qualifying and address concerns
Generative remarketing
Contextual UI Fraud detection AI Systems to dynamically
Adaptive user interfaces, by AI for analyzing customer build cohort-based and
context: behavioral, social, behavior to identify and personalized remarketing
temporal, emotional, respond to fraud or suggest strategies
personal new systems that might
further prevent fraud
Conversational UI
Users use natural language
(voice or text) to engage,
filter, qualify, assist, etc.
during buying experience

Carousal Starts

Responsible AI
Unlike the software solutions of the pre-generative AI
world, generative solutions cannot be built, tested, and
released into an ecosystem without continuous
oversight. Continuous governance is mandatory, and
it’s cultural.

Carousal Ends

Deploying gen AI in a responsible and effective way

With all of the compelling use-cases for gen AI and the immediate accessibility of public tools in the market today, it can be easy
to get carried away in the AI hype. That same consumer availability of basic AI tooling can trivialize the complexity and
downplay the policy, process, partnership and skill required to build tailored, production-grade solutions.

Enterprise AI and machine-learning (ML) applications require automation engineering, data management, feature engineering,
resource management, QA, and testing teams, IT operations and more; all aligned with coherent, shared accountability in order
to effectively deliver solutions. In the following pages, we will explore how LLMOps expands our view of DevOps and how an
updated view of quality engineering can safeguard AI solutions with holistic automated testing.

It all starts with setting strong, enforceable principles for responsible AI development.

Guiding principles for responsible AI development

Tactical principles for developing gen AI solutions in efficient, safe and value-oriented ways

Be robust, be safe Design for transparency

AI systems should perform reliably and safely. By building AI systems should be understandable. Build trust and drive
and deploying AI in accordance with best practices where understanding through silo-breaking collaboration and rich
we robustly test before deployment then monitor and communication across users and stakeholders, allowing
improve operations regularly, we can reduce the risk of them to understand AI systems and system outputs within
harm or unintended outcomes. their own, personal context.

Benefit people and communities Promote inclusivity and minimize bias

Build a more sustainable and inclusive world through AI Bias exists in our data, models and our world; responsible AI
innovation. AI outcomes must incorporate human benefit systems seek to ensure AI is fair, unbiased and
and environmental sustainability in order to deliver impact representative end to end and full-context. AI systems
and value to shareholders, users, customers, employees should treat people fairly and AI should be produced and
and society at large. reviewed by diverse teams.

Protect privacy and respect boundaries Drive accountability and enable participation

AI systems must be secure, compliant and respectful of People should be accountable and in control of AI systems.
people. Affirmative consent and a human-centered, Clear processes and incentives for engagement create a
privacy-first approach ensures sensitive data is never used culture where every individual is empowered to protect
unethically. A variety of auditing systems and safeguards people, minimize risk and discover spaces of humane
are key within gen AI systems. value.

LLMOps brings speed, support and safeguards to solution

development
The following two pages provide an introduction to LLMOps but remain too high-level to sufficiently detail the orchestration of
people, tooling and processes required to operationalize these practices. To explore more, engage with Cognizant’s generative
AI team.

Gen AI solutions can be surprisingly seamless. The ability to understand users, act on their needs and provide human-like
creative responses is what makes gen AI such a compelling solution today. Behind the scenes, though, gen AI solution
development adds layers of complexity to the work of digital teams that go well beyond API keys and prompts. Orchestration is
critical.

Consider the following difficulties:

• Gen AI systems are difficult to test and unpredictable by nature – they cannot be validated and deployed in a ”set and
forget” fashion

• Development requires many skills and teams must rely on each other – this takes coherent organization and alignment
programs

• Gen AI systems must be at once predictable but flexible – this takes training and continuous monitoring

• Generative AI development requires too many skills and processes to manage with manpower alone

• Generative AI often leads to scenarios where privacy, regulatory compliance and data leakage are concerns—systems must
be secure

• Manual review of issues, anomalies and experience is untenable and might remove any ROI or advantage of these systems—
automation is needed

Due to these conditions, new ways of working are required. Building on the popular concept of DevOps, LLMOps provides a path
forward.

LLMOps defined: LLMOps aligns data engineering, agent development, software engineering and IT Operations to enable
continuous integration, continuous delivery and continuous model training with an emphasis on automation and monitoring at
all steps.

As organizations seek to develop effective generative AI- enabled solutions for internal and external users, defining and
enforcing their own LLMOps approach is imperative.

This often starts with defining the KPIs of gen AI solutions (aligned to responsible AI principles) and ensuring that processes,
governance and tooling are in place—made possible by LLMOps—to monitor and influence those KPIs.

These LLMOps KPIs can include:

Cycle time: Deploy frequency: Review time: Automated Data and model
Duration from How often updates Time taken to solution ratio: drift:
initiation to or new gen AI review and The proportion of Discrepancies in
deployment of gen solutions are validate solution tasks handled solution (model)
AI-based solutions pushed to (model) outputs autonomously by outputs or
production and performance the system incoming data
from the training
scenario

LLMOps and AIOps are often confused, but they describe entirely different disciplines: LLMOps standardizes machine
learning model deployment while AIOps automates IT operations.

Transparency and automation at the heart of LLMOps

As organizations tiptoe into gen AI, linear solution development processes will be
favorable for proof-of-concept development at speed. The belief is that model
training is something done early within a process and that a trained model can be
utilized endlessly. This can work for testing the waters but is not a sustainable
“We must build quality
solution. and control into AI
solutions to manage
Mature LLMOps processes are iterative in nature with observability and
automation at their heart. As a continuous cycle, LLMOps allows data intake and
their continuous
learning to regularly impact the solution while automating as much as possible evolution. Due to their
and keeping humans in the loop. This feedback loop is key to responsible AI broad capabilities
development. By ensuring that model behavior, application performance, data and emergent
protection and system changes are controlled through a technology-driven
behavior,
workflow, organizations can operate more effectively.
management is
needed across the
entire lifecycle.”

Andreas Golze

Head of Quality Engineering &

Assurance Practice

Quality assurance and testing systems in a gen AI world

The evolved role of quality assurance’s (QA) teams and tooling within the delivery process will be a critical focus area for
organizations seeking to deploy LLMOps.

While QA automation has become an area of strength for many mature engineering organizations, traditional approaches
are insufficient for generative AI. The scope of QA and test automation has changed, with new driving factors to consider for
AI-based applications.

Updated QA automation capabilities must cover:

• Systemic data assurance: Supports operations and decisions across process
e.g., Feature selection analysis, input data augmentation, data fitment analysis

• Systemic model assurance: Automation for monitoring model behaviors

e.g., Model fairness analyzer, model validator, model interpreter, adversarial robustness tools

• Generative output assurance: Automation to monitor outputs of gen AI

e.g., naturalness, toxicity, coherence, demo polarity, clarity, truthfulness, relevance, etc.

• Low data scientist QA efforts: Services and tooling to reduce specialist effort
e.g., low-code/no-code solutions, customizable observability UI, etc.

• Deep insights: Automation of insights from AI solutions / workflows

e.g., gen AI tooling to explain solutions, model outputs, live data, etc.

Carousal Starts

Readiness
Preparing an organization for the generative AI world
doesn’t mean moonshots and risk, it means foundation
laying and learning. Setting up governance now will
pay dividends near and long term.

Carousal Ends

Practical guidance to reap the benefits of gen AI

Despite the hype around gen AI, we’re still in the early days of the AI-driven business. It’s a certainty that AI will transform every
corner of our digital universe and yet we’re continuing to learn how. With new applications conceived daily and development of
next-gen generative AI models underway, innovators are fast at work reshaping the future of work. Adaptability in such a rapidly
changing landscape is critical.

With so much hype and a sea of noise to cut through, many organizations are asking more tactical questions. What must be
navigated to move forward?

To get practical about gen AI, start with these questions. In the following pages, we’ll double-click into each as we explore a path
forward.

“To get practical about gen AI, have we:”

1 2 3 4 5
Provisioned initial Designed a path Introduced AI to the Modernized and Set new expectations
access to to scale development process? significantly automated with our suppliers?
enterprise-grade successful POCs? core business operations?
gen AI tools?

Provisioning initial access to enterprise-grade gen AI tools

In the wake of ChatGPT’s emergence, it’s safe to say that every enterprise was abuzz
with cautious excitement about the potential of this new technology. Steps to take

At the individual team-member level, workers around the world began testing • Update policies and governance
generative AI for their own use-cases. A recent survey from The Conference Board A blocker for many orgs today, set
found that 56% of workers are using gen AI on the job, with nearly one in 10 clear policies and governance for
employing the technology on a daily basis. Yet just 26% of respondents said their a generative AI world
organization has a policy related to the use of generative AI, with another 23%
reporting such a policy was under development.6 • Educate employees actively,
communicate often
Organizations have been relatively quick to respond to these risks. Amazon, notably, Clear communication to
found7 indication of proprietary data in public model usage and responded with a associates around access, policy
ban. leadership, support and
appropriate usage
This isn’t luddite behavior, it’s just good practice. At this early stage, it’s unclear
exactly how customer data, proprietary business data and other protected data is • Set governance over approved
either being exposed to the operators of public LLMs or used to train the models services
themselves. Couple this with the simpler considerations of Privacy Policy adherence, Clarify approved gen AI solutions
Terms of Service, regulatory considerations and more bans are surely on the horizon. and establish ownership for
access provisioning and support
But still, the advantages…

As new products go, any amount of friction (cost, risk, etc.) can have a chilling effect • Identify and supervise adoption
Maintain observability over
on adoption. But generative AI isn’t simply a new product; it’s a transformative
approved AI services, monitor
technology that can change the world in striking, progressive ways.
who is using them and for what
Early adopters will have the advantage in this new world. Beyond the obvious purposes
cultural and process execution benefits of gen AI, we expect a patent boom in the
coming years as organizations invent novel uses of gen AI-based tools within their • Learn and capture best-practices
business. Source insights and best
practices within an organizational
Preparing the business for gen AI means getting serious about near-term, safe- context, communicate learning
guarded adoption with well-integrated monitors and control of usage. Gain
advantage while minimizing risk and learn as you go.

Design a path to scale successful POCs

We’re quite early in generative AI’s diffusion of innovation bell-curve. Early adopters
are establishing and quantifying basic use cases—gaining earned media as a Outcomes in focus
result—and most would-be digital leaders are watching with curiosity.
• Build a business case for
expanded focus
Give enterprise context to gen AI
use-cases by showing viability
and capability against revenue-
driving focus areas, expand on
success.

• Gain edge, protection from

disruption
Leverage gen AI early for
disruptive use-cases that drive
business value, safeguard the
business and accelerate future
In an August 2023 report by Bain and Company8, only 6% of surveyed health projects.
systems executives have a clear gen AI strategy in place, yet 75% believe that the
technology can reshape their industries. The reasons are the same that we’ve • Develop early internal SMEs and
already discussed: uncertainty, risk, lack of inside knowledge and indecision. This insight
lull in early adoption is where the advantage lies. Begin establishing subject matter
expertise within core internal
Quietly building a boom teams by capturing and sharing
knowledge during practical
To ready themselves for the road ahead, it is imperative that organizations go application.
beyond provisioning access to public tooling and begin developing their own
inside use-cases to drive a business case, spark thinking and lay a foundation for • Build internal and external
future development. This can be done in phased, controlled and protective ways. reputation
Establish industry point of view on
We suggest two complimentary approaches:
gen AI, attract business and
1. Establish and run pilot projects talent attention and set a cultural
foundation for an AI enterprise.
Pilot projects build a ‘light house’ for future innovation and expansion. By
establishing specific initial goals for a cross-functional pilot project team to pursue,
organizations can create disruptive proofs of concept and establish an internal
POV.

2. Enable and accelerate with AI innovation “labs”

We have supported multiple organizations on establishing their own innovation

lab environments where governance, collaboration and technology enablement
are high. These environments become particularly powerful when formed in
collaboration with hyperscalers who might provide innovative organizations with
access to advanced models, education and specialized tooling.

Identify opportunities to add AI to the development process

As covered in our section on LLMOps, generative AI development implies
systemic changes to the way that software is delivered and supported
within organizations. Zooming in, the same goes for software product
engineering processes.

Here are just some ways that gen AI might impact product development.

Idea generation
The ability of Generative AI applications to work with trained models
while evolving those models (and the application’s outputs) with the
consumption of real-time data can unlock compelling use-cases for
product idea-generation. Rather than relying on surveys and user reviews
for qualitative data, Generative AI agents might deliver new concepts
frequently based on real-time analytics. Product managers can then link
these ideas to business goals and set a path forward.

Product design
As multimodal models (capable of intaking and outputting images, text,
audio, etc.) mature and see enterprise adoption, “clickable prototype”
design will become less a job for designers and instead be handled by
gen AI tools. Fed with design principles, systems and reference designs,
these prototype design tools will produce unbiased prototypes best fitting
the market data available. The job of designers will be to identify the most
promising solutions and refine them.

Risk mitigation
A core responsibility in product management is to manage and mitigate
risk. With its predictive analytics capabilities, AI tooling can help in
identifying potential risks and roadblocks early on in the prototyping
phase. Quality, market readiness and future success can all be gauged
by having algorithms analyze historic data, user preferences and even
real-time market trends.

Resource optimization
Sustainability is the challenge of this generation of business. Generative
AI can support sustainability efforts by optimizing resources and material
mix for minimized waste and environmental friendliness. It can take
regulatory processes into account, report on data and even affect
subsequent production processes for both software and physical goods.

Modernize and significantly automate core business

operations
There is no doubt that gen AI will revolutionize how we work. Process automation has long been a popular use-case in our
digital world and AI is going to open entire new opportunity spaces here. The debate around automation will continue to be
more focused on how regulators will impose limitations on the technology instead of how much potential the technology
affords us.

Like media streaming, ride-sharing and other tech-driven economic catalysts that regulators have tried to stay in front of,
resistance is a bit futile. Technology will be squeezed for as much juice as it can offer in a free market.

In our opening section of this document covering the future of gen AI, we touched on a shift from a war for talent (commonly
discussed in the 2010s and pandemic era) towards a war for innovation as all businesses use gen AI to gain efficiency. AI-
enabled automation of core business processes will drive this shift. Savings must be reallocated to accelerating business
value.

1. Target core functions 2. Assess and expand on 3. ID opportunities to pre- 4. Deploy generative
for AI-driven existing process process data assurance
augmentation automation
Utilize gen AI to reduce Apply gen AI to business
Explore areas for Apply gen AI solutions to or replace human data processes to
generative AI to support existing process processing, gaining continuously and
knowledge work and automations to further speed and accuracy completely audit quality,
accelerate creativity streamline for efficiency regulatory compliance
and more

By 2032, few jobs could go untouched by generative AI

Some impact: Greatly impacted: Little impact:

Exposure scores of at Exposure scores of at Exposure scores of 5%
least 5% least 25% or lower

Most jobs will see some change from gen AI, and over half could be greatly impacted.

Set new expectations with your suppliers

It’s every company’s job to evolve
From supplier to partner
The early big press-makers of generative AI have been the expected parties.
Hyperscalers have introduced new or evolved platforms for building AI solutions As the innovation potential of
within their ecosystems. Myriad ultra-specialized startups have announced generative AI becomes clear to more
compelling new solutions to old problems (e.g., Hyfe’s10 cough sound monitoring for organizations, the opportunity to
illness diagnosis). And service providers, like us, are launching new accelerators and create wholly new experiences,
labs for gen AI development. services and processes by
partnering with suppliers on a joint
But generative AI is coming for every product, in every market. That goes well journey will become compelling for
beyond the early adopters and it includes you, your competition and your suppliers. many businesses.

As you seek to leverage gen AI to unlock new efficiency, differentiate experiences, Going well beyond the cost savings
maximize quality, find cost-savings and evolve the business model, don’t discount of a joint investment, with enriched
the role your suppliers will play in these improvements. data, access to more skills and
beyond, these partnerships might
Whether a service provider, a manufacture or raw goods provider, a logistics service benefit both parties in dramatic
or any other company that plays a role in your operations, there is an advantage to ways when executed well. Consider
engaging early in a dialogue about gen AI. the role of each key supplier within
your service or product delivery and
Every one of your suppliers will be at a different stage of this journey. As they
move the discussion beyond what
navigate use-cases, seek to answer questions about risks and control and otherwise
they can do with AI for you.
dive into gen AI, join them.

Why engage now?

You stand to gain from their You will learn from one another Open the door to more data
improvements
As all companies are learning, work As you engage with your suppliers,
Suppliers are critical to your bottom with suppliers to understand their consider internal solution
line. Ask how they plan to improve own findings, partnerships and opportunities and how supplier data
SLAs, decrease total cost of interest areas. Formalize shared might improve model training and
ownership, operate faster and learning between strategic teams solution delivery. Bring predictability,
otherwise drive more business value and capture insight. new product ideas, streamlined
for you and other customers. operations and more to life.

To learn more about Cognizant’s generative AI solutions:

https://www.cognizant.com/us/en/services/ai/generative-ai

Dive deeper into this topic and learn about the kinds of gen
AI work Cognizant is doing with our clients:
https://www.cognizant.com/us/en/insights/tech-to-
watch/generative-ai

References
1. https://www.cnn.com/2022/07/23/business/google-ai-engineer-fired-sentient/index.html cognizant-technology-to-invest-1-billion-in-gen-ai-over-three-years-
123080300490_1.html

2. https://www.crunchbase.com/hub/artificial-intelligence-startups

3. https://thehill.com/policy/technology/3954570-google-ceo-says-ai-will-impact-every-product-of-every-company-calls-for-regs/

4. https://tvpworld.com/71221233/ai-threatens-nearly-30-of-jobs-within-oecd-report

5. https://devclass.com/2023/02/16/github-claims-new-smarter-copilot-will-block-insecure-code-writes-40-60-of-developer-output/

6. https://www.conference-board.org/press/us-workers-and-generative-ai

7. https://gizmodo.com/amazon-chatgpt-ai-software-job-coding-1850034383

8. https://www.bain.com/about/media-center/press-releases/2023/majority-of-health-system-executives-believe-generative-ai-will-reshape-the-industry-yet-only-6-have-
a-strategy-in-place/

9. https://www.cognizant.com/us/en/gen-ai-economic-model-oxford-economics

10. https://www.hyfe.ai/

Sitemap Terms Privacy Notice Cookie Notice ©2024 Cognizant, all rights reserved